Multi-language MCP server repository with Node.js smart home light control via Matter protocol and Python stock price information service
This Stock MCP server has fundamental definition quality issues. While it provides 5 tools/prompts with generally reasonable naming (get_current_stock_price, get_historical_stock_price follow verb_noun convention), the definitions lack critical elements required for production LLM agent use. Tool descriptions are present but generic ('Tool function for getting current stock price information'). Parameter descriptions are minimal. No output schemas are documented anywhere in the visible code. The tools appear to work but lack the rigor expected of production-grade agent toolkits. Critical issue: prompts 3-5 (stock_current_price, stock_historical_price, stock_usage_guide) are labeled as 'prompts' in the server output but appear to be callable resources/tools in the code, this is a categorization mismatch that violates MCP semantics.
Gets current stock price information for a given ticker symbol
Gets historical End-of-Day (EOD) stock price information for a specific date
Gets the current state of the smart light. Returns whether the light is currently ON or OFF. Use this when the user asks about the light's status or wants to know if it's on.
Toggles the smart light state. If ON, turns it OFF. If OFF, turns it ON. Use this when the user wants to switch or toggle the light without specifying a specific state.
Turns the smart light OFF. Use this when the user wants to turn off, disable, or deactivate the light.
Turns the smart light ON. Use this when the user wants to turn on, enable, or activate the light.
No output schemas documented for any tool. The code fetches stock data via TwelveData API and formats it via format_data(), but the server definition provides zero documentation of what fields the LLM should expect (e.g., current_price, change, high, low, volume).
Tool descriptions are generic and do not guide LLM decision-making. 'Tool function for getting current stock price information' does not answer: WHEN to use this vs. a similar tool? WHAT does it return? WHY would an agent select this? Per pattern:tool-description, descriptions must state WHAT, WHEN, and CONSEQUENCES. Rubric baseline: 'Average tool description length: 194 chars (p10=34, p90=392)', these are all ~50 chars, incomplete.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 9 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 42 | - | v1 |
Parameter 'symbol' is described only as 'Stock symbol (e.g., AAPL, MSFT, GOOGL)'. LLMs tend to reuse example values literally rather than adapting to the actual context.' The description should instead state the format: 'Uppercase stock ticker symbol (1-5 uppercase letters)'.
Parameter 'date' in get_historical_stock_price is described as 'Date in YYYY-MM-DD format' but lacks explicit validation guidance. The implementation checks for weekend dates and returns a warning, but the parameter description does not document this behavior.
Error handling exists in the code (_fetch_current_stock_price, _fetch_historical_stock_price) but is not documented in the tool definitions. Errors return strings like 'Error: No EOD data returned' which are informative but lack the pattern:recovery-guide. The LLM does not know if it should retry, ask the user, or try a different tool. Per pattern:error-classification, errors must be categorized as retryable, user-fixable, or fatal.
The five tools listed include both tool definitions (get_current_stock_price, get_historical_stock_price) AND prompt definitions (stock_current_price, stock_historical_price, stock_usage_guide). Per MCP semantics, prompts are SERVER-INITIATED INSTRUCTIONS sent to the LLM, not client-invoked tools. The file shows @mcp.prompt decorators for stock_current_price and stock_historical_price, confirming they are prompts, not tools. The server description conflates them. This is a semantic violation of the MCP protocol model.
No input validation constraints documented. Symbol parameter accepts any string, no enum, no regex, no length limits. The code calls twelve_data_client.quote(symbol=symbol.upper()), which relies on TwelveData API to reject invalid symbols. Per pattern:constrained-input, parameters should declare valid values to allow LLMs to select correct inputs upfront rather than fail at runtime.